Documente Academic
Documente Profesional
Documente Cultură
(hazard function)
fT (t)
h(t) = (2)
1 − FT (t)
FT (t) = 1− e−H(t)
t (4)
H(t) = ∫ h(u)du
0
To prove Eq. 4, let G(t) = 1 - FT(t). Then G’(t) = dG(t)/dt = - fT(t) and, from Eq. 2, -h(t)
= G’(t)/G(t). Integration of both sides from 0 to t gives
t t
G' (u)
− ∫ h(u)du = ∫ du = ln G(t) − lnG(0) (5)
G(u)
0 0
t
Since ∫0 h(u)du = H(t) and G(0) = 1 (hence lnG(0) = 0), one concludes from Eq. 5 that
−H(t)
lnG(t) = - H(t). Therefore G(t) = e , which implies FT (t) = 1− e−H(t) in Eq. 4.
= P[T ≤ t + t o T > t o ]
P[(T ≤ t + t o ) ∩ (T > t o )]
= (6)
P[T > t o ]
e −λto − e −λ (t + to )
=
e −λt o
= 1 − e −λt = FT (t)
Eq. 6 shows that the distribution of the remaining life is the same as the original
distribution, i.e. the same as the distribution for a new item.
Items that are worse old than new have lifetime distributions with increasing hazard
functions and items that are better old than new have decreasing hazard functions. Notice
however that the hazard function needs not be monotonically increasing or decreasing; it
may also fluctuate. For example, many new manufactured items go through a “break-in”
period in which failure is relatively likely to occur due to manufacturing defects
(“lemons”). For a period following this initial break-in phase, failure occurs much less
frequently. However, after a while, the incidence of failures increases, due to factors like
wear, aging and fatigue. For such items, the hazard function has a characteristic
“bathtub” shape, as the one shown below (incidentally, such bathtub hazard functions are
appropriate also for human life!).
Figure 1: Schematic hazard function for items with break-in and aging.
Problem 8.1
(a) You are about to buy a used car, which has accumulated 50,000 miles without major
problems. You are worried that something bad may soon be happening to it. The
dealer reassures you that the car “is very safe, because statistics show that its
lifetime distribution is uniform between 0 and 150,000 miles.” In his sale pitch, the
dealer adds: “ This means that the car has the same chance of dying on you in the
next 50,000 miles as it did during the first 50,000 miles after rolling out of the
factory. Therefore, it is as good as new.” Using Eq. 2, calculate the hazard function
h(t) for the car when it was new (notice that here t has the units of miles). Compare
h(t) for 0 < t < 50,000 miles and 50,000 < t < 100,000 miles. Do you agree with the
dealer that, for the next 50,000 miles, the car is “as good as new?” Explain your
reasoning.
(b) Now suppose that the dealer tells you that, for t > 50,000 miles, “cars like this have a
lifetime distribution with probability density function fT(t) = t-a, for some a > 1.”
Calculate and plot the hazard function h(t) for a = 2 and t > 50,000 miles. Does this
function increase or decrease with t? Is an older car of this type more or less reliable
than a newer car?